Rapid Optical Imaging of Breast Tumor Margins: Final Results from a 100-Patient Clinical Study.

Abstract

Abstract Background: Increasing attention has been given to the surgical margin status in primary breast conserving therapy (BCT) due to reports which indicate that 20-70% of patients undergoing BCT must undergo multiple surgeries for complete tumor resection. For the past 4 years, our multidisciplinary group has been working to address this clinical need via the development of an optically-based intraoperative breast tumor margin assessment device. This device, based on diffuse reflectance spectroscopy, is sensitive to biochemical and morphological changes associated with cancer and has a sensing depth of 1-2 mm which is compatible with widely used criteria for cancer-free margins. The device is capable of surveying the entire specimen surface intraoperatively in imaging mode. Here we report the results of a non-significant-risk study of the device in over 100 patients at Duke University Medical Center.Materials and Methods: Under an institutionally-approved protocol, we tested our device on consented patients undergoing a partial mastectomy at DukeUniversity Medical Center for invasive or in situ malignancies. Optical spectral images of tumor margins were recorded within 20 minutes of excision and converted into tissue compositional parameter maps that reflect the vascular density, fat content and cell density in the tissue. The pathologic status of the imaged margins was collected from standard post-operative surgical pathology reports. Intraoperative frozen section and touch prep analysis was not performed on these specimens. Margins were considered positive if residual malignancy was found within 2 mm of the tissue surface.Results and Discussion: BCT specimens from 121 patients have been imaged with the optical device. One to four margins were imaged on each BCT specimen. Data from 112 patients were retained for analysis (51 negative margins, and 47 margins containing cancer within 1mm). Images of negative margins from patients with at least one positive margin (9 patients) elsewhere on the specimen were excluded from analysis due to the potential for presence of margin positivity that was not identified pathologically. For each margin, a set of 8 tissue composition maps were generated, from which a set of 36 image-descriptive variables were obtained. Wilcoxon rank-sum tests were used to determine which of the variables best separated negative from positive margins. A predictive model was developed using conditional inference trees to identify the optimal partitions from all 36 image-descriptive variables. The model selected variables related to the light scattering properties, total hemoglobin content, and β-carotene content of the underlying tissue, which are related to tissue density and morphology, vascular volume, and fat content, respectively. This model resulted in a sensitivity of 80% for detecting cancer at the margin, a sensitivity of 73% for detecting residual disease within 1mm of the margin, and a specificity of 65%. These results are promising and a prospective validation trial of the device is under development. Citation Information: Cancer Res 2009;69(24 Suppl):Abstract nr 5017.